memex:steward:dashboard
Launch a loopback-only, read-only HTTP dashboard over Memex — per-store summary, the federated index, knowledge communities, the code-navigation graph, and the agent registry, plus keyword search, a click-to-read content overlay, and an interactive 3D knowledge-graph view.
- License 许可
- MIT
- 所属仓库
- nitekeeper/memex
- 最近更新
- 2026-09-22
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- memex:steward:dashboard
- description
- Launch a loopback-only, read-only HTTP dashboard over Memex — per-store summary, the federated index, knowledge communities, the code-navigation graph, and the agent registry, plus keyword search, a click-to-read content overlay, and an interactive 3D knowledge-graph view.
03 SKILL.md 正文
memex:steward:dashboard
A read-only observability surface the Data Steward exposes: it opens every
registered store (plus the fixed-path code_graph.db) read-only, aggregates
a cross-store summary, and serves a single self-contained HTML page over a
loopback-only stdlib HTTP server. It writes to nothing, so it is not a
Librarian write-path bypass (spec §6 / M3 govern document writes; this is pure
read-side reporting).
When to use
- The user asks to see / visualize / get an overview / dashboard of what is stored in Memex.
- A quick health glance before/after a large ingest, audit, or release.
- For a deep integrity audit (orphans, broken relations, schema drift) use
memex:steward:auditinstead — this skill is a friendly overview, not an integrity checker.
Inputs
--port— bind port (default8765; auto-increments if busy, up to +20).--host— bind host (default127.0.0.1). Non-loopback hosts require--allow-non-local(the dashboard exposes your Memex contents).--no-open— do not auto-open a browser (use in headless/SSH sessions).--once— print the summary JSON to stdout and exit (no server). Use this for scripting, smoke checks, or when a browser is unavailable.
Invocation
The implementation is scripts/dashboard.py (stdlib only — http.server +
sqlite3 + json; no third-party deps). Run it as a module with the plugin
root on PYTHONPATH. <PLUGIN_ROOT> is the path resolved in memex:run Step 0.2.
Serve the dashboard (long-running — start it in the background and report the
URL to the user; it prints Memex dashboard → http://127.0.0.1:<port>/):
PYTHONPATH="<PLUGIN_ROOT>" python3 -m scripts.dashboard --no-open --port 8765
Then tell the user to open the printed http://127.0.0.1:<port>/ URL. Stop it
with Ctrl-C (or kill the background job).
One-shot JSON snapshot (no server — for a quick textual summary):
PYTHONPATH="<PLUGIN_ROOT>" python3 -m scripts.dashboard --once
Routes served
| Route | Returns |
|---|---|
GET / |
the self-contained dashboard page (keyword search box + a ◉ 3D graph link). Deep links: ?q=<keyword> prefills+runs the search, ?doc=<index_id> opens a document. |
GET /graph |
an interactive 3D knowledge-graph viewer (Obsidian-style) — documents as nodes, relations as edges, colored by community; orbit / zoom / search / click a node to open its content. Dependency-free vanilla JS + canvas (no Three.js/CDN, under the same CSP). |
GET /api/summary |
the cross-store summary as JSON (recomputed live each request) |
GET /api/graph |
the index knowledge graph as JSON {nodes, links, truncated} (read-only; dangling/self edges dropped; capped at 900 nodes) |
GET /api/search?q= |
keyword search across three surfaces — documents (federated index, FTS5 + LIKE fallback), the agents registry (roles + agents by name/description/profile), and the code graph (node labels per repo) — each result carries a kind (document/agent/code); the UI groups by kind |
GET /api/doc?id= |
detail for a clicked result, dispatched by id prefix: a bare index_id → document content from its source store; role:/agent: → a registry profile; code:<repo>|<node> → a code-graph node summary (location + callers/uses) → {title, subtitle, content, …} |
GET /healthz |
{"ok":true} liveness probe |
Both the dashboard search list and a 3D-graph node click open a shared, read-only content overlay. Document bodies render as Markdown via a dependency-free DOM builder (createElement/textContent, sanitized hrefs); raw/rendered toggle.
Safety contract
- Read-only. Every store is opened with SQLite
mode=ro; no write path. - Loopback by default. Binds
127.0.0.1; refuses a non-local bind without--allow-non-local. - No filesystem serving. Only the three fixed routes above respond; there is no static-file handler, so no path-traversal surface.
- No untrusted templating. The page carries zero server-side interpolation;
all data crosses as JSON and is rendered client-side via
textContent, so a stored document title containing markup cannot become script (CSP also set). - Degrades gracefully. A missing store/table/column renders as empty rather
than erroring — a freshly bootstrapped install (no
code_graph.db) is fine.
04 同仓库更多 Skills
run
Use to invoke any Memex v2.0 operation. Routes user-facing intents (ingest, ask, capture, lint, synthesize, dashboard) and agent-facing CRUD primitives to the right internal procedure.
memex:steward:reconcile-orphan
Authorized fix-up of a flagged orphan. Four actions — delete-index (remove documents row), repair (backfill row_id when target exists), reindex (re-run Librarian; deferred), note (acknowledge). Raises OrphanNotFoundError when index_id is unknown or, for repair, already linked.
memex:steward:audit
Run a full integrity audit across the Memex Index and all registered stores. Detects orphans, broken relations, schema drift. Writes a structured audit report to ~/.memex/audits/. Read-only — never auto-fixes.
memex:steward:audit-store
Run an integrity audit scoped to a single registered store (reverse orphans + schema drift). Lighter than full audit.
memex:index:write
Direct write to the federated Index — the lower-level primitive that memex:brain:ingest and memex:brain:capture build on. Use when a consumer (Atelier, custom plugin) needs to add a document to its own store and have it indexed by the Librarian, without going through the Brain layer. Routes through Librarian subagent + Memex Core.
memex:index:search
Federated search across every registered Memex store via the Reference Librarian. Lower-level primitive that memex:brain:ask builds on — use directly when a consumer (Atelier, custom plugin) needs raw query results without Brain-specific formatting.